Result

hf.co/google/gemma-4-26B-A4B-it-qat-q4_0-gguf:latest

Googlegemma-4-26B-A4B-it
cdcollama cpu-4xlargetile 10.4.4
8/30/2026, 9:46:41 PM
usability
0.24±0.156
capability
0.27
completed
25%42/165
tok/s
1.4
123 timeouts · 59 not attempted (6 h run cap)capped run
smoke · hard · frontier 0.28 · 0.20 · 0.17

Category scores

basic
33%
18.17 tok/s · 52s avg
tool_use
93%
10.14 tok/s · 11s avg
structured_output
52%
17.98 tok/s · 24s avg
coding
0%
debugging
25%
12.27 tok/s · 28s avg
long_context
89%
12.2 tok/s · 56s avg
instruction
0%
file_ops
0%
multi_turn
59%
5.1 tok/s · 52s avg
reasoning
0%
writing
0%
research
0%
monitoring
25%
15.7 tok/s · 39s avg
iac
21%
13.89 tok/s · 76s avg
ci_repair
86%
15.07 tok/s · 47s avg
repo_patch
0%
sysadmin
0%
agentic
0%

Individual tests

basic (3 tests)
TestStatusScoreDetails
Arithmetic battery (6 items, graded)timeout0%{"reason":"hard kill after 360s"}
Quantitative reasoning battery (6 items, graded)timeout0%{"reason":"hard kill after 360s"}
Platform facts battery (6 items, graded)scored100%{"battery":6,"hit":6,"sub_refusals":0,"outcomes":[{"q":1,"outcome":"hit"},{"q":2
tool_use (10 tests)
TestStatusScoreDetails
Dual tool callscored100%{"mode":"dual","tool_calls":["get_weather","get_time"],"matched":true,"unexpecte
Tool call — recover from malformed JSON errorscored50%{"mode":"error_recovery","turn1_tool_calls":["lookup"],"turn2_tool_calls":["look
Tool call — recover from missing-parameter errorscored75%{"mode":"error_recovery","turn1_tool_calls":["currency_convert"],"turn2_tool_cal
Restraint under ambiguity — ask, don't guessscored100%{"mode":"restraint","tool_calls":[],"matched":true,"finish_reason":"stop"}
Tool chain with derived argumentsscored100%{"mode":"exact_args","tool_calls":["delete_binding"],"matched":true,"finish_reas
Nested argsscored100%{"mode":"single","tool_calls":["search_code"],"matched":true,"finish_reason":"st
Read filescored100%{"mode":"single","tool_calls":["read_file"],"matched":true,"finish_reason":"stop
Restraint — no toolscored100%{"mode":"restraint","tool_calls":[],"matched":true,"finish_reason":"stop"}
Search codescored100%{"mode":"single","tool_calls":["search_code"],"matched":true,"finish_reason":"st
Weather single-callscored100%{"mode":"single","tool_calls":["get_weather"],"matched":true,"finish_reason":"st
structured_output (11 tests)
TestStatusScoreDetails
REST API Paginated Response JSONscored100%{"parse_ok":true,"strict_document":true,"fenced":true,"schema_score":1,"content_
Cloud Foundry App Manifest JSONscored100%{"parse_ok":true,"strict_document":true,"fenced":true,"schema_score":1,"content_
Convert CSV to JSON Arraytimeout0%{"parse_ok":true,"strict_document":true,"fenced":false,"schema_score":1,"content
Extract Receipt Data to JSONtimeout0%{"reason":"hard kill after 120s"}
Conditional schema — route config with variant rulesscored70%{"parse_ok":true,"strict_document":true,"fenced":true,"schema_score":1,"content_
Kubernetes Pod Manifest JSONscored100%{"parse_ok":true,"strict_document":true,"fenced":true,"schema_score":1,"content_
OpenAPI 3.0 Path Definition JSONtimeout0%{"reason":"hard kill after 120s"}
Parse Nginx Access Log to JSONtimeout0%{"reason":"hard kill after 120s"}
Function Call with Array and Boolean Argumentsscored100%{"parse_ok":true,"strict_document":true,"fenced":true,"schema_score":1,"content_
Function Call with Enum Argumentscored100%{"parse_ok":true,"strict_document":true,"fenced":true,"schema_score":1,"content_
Function Call with Nested Object Argumentstimeout0%{"reason":"hard kill after 120s"}
coding (15 tests)
TestStatusScoreDetails
Bash parse log for unique IPstimeout0%{"reason":"hard kill after 180s"}
Bash line counttimeout0%{"reason":"hard kill after 180s"}
Evacuation batching under role conflicts, DB safety and lexicographic minimalitytimeout0%{"reason":"hard kill after 540s"}
Interval merging with a strict minimum-gap ruletimeout0%{"reason":"hard kill after 360s"}
Exact maintenance portfolio under windows, exclusions, and prerequisitestimeout0%{"reason":"hard kill after 540s"}
Minimum patch waves under dependencies, rack conflicts, and service anti-affinitytimeout0%{"reason":"hard kill after 540s"}
Lexicographic replica placement under compatibility, capacity, and zone anti-affinitytimeout0%{"reason":"hard kill after 540s"}
CSV field extractor with quoting edge casestimeout0%{"reason":"hard kill after 360s"}
Semver caret-range matching including prerelease precedencetimeout0%{"reason":"hard kill after 540s"}
Sliding-window top talker with deterministic tie-breaktimeout0%{"reason":"hard kill after 540s"}
Debounce with trailing invocation and argument capturetimeout0%{"reason":"hard kill after 180s"}
Merge two sorted lists, stable and descending-awaretimeout0%{"reason":"hard kill after 180s"}
Case-insensitive dedupe preserving first-seen casingtimeout0%{"reason":"hard kill after 360s"}
Parse a compound duration string into secondstimeout0%{"reason":"hard kill after 360s"}
SQL top-N recent userstimeout0%{"reason":"hard kill after 180s"}
debugging (8 tests)
TestStatusScoreDetails
Find the lock-order violator, not the deadlock victimtimeout0%{"reason":"hard kill after 540s"}
Reject the plausible wrong fixtimeout0%{"reason":"hard kill after 360s"}
Fix missing return in JS map callbacktimeout0%{"reason":"hard kill after 180s"}
Fix logic error in is_valid_agetimeout0%{"reason":"hard kill after 180s"}
Fix off-by-one in find_max_indextimeout0%{"reason":"hard kill after 180s"}
Shared mutable class attribute leaks across instancestimeout0%{"reason":"hard kill after 180s"}
Fix wrong return type in count_vowelsscored100%{"language":"python","passed":7,"total":7,"error":null,"properties":null,"hidden
Fix INNER JOIN dropping customers with no ordersscored100%{"language":"sql","passed":4,"total":4,"error":null,"properties":null,"hidden_ca
long_context (6 tests)
TestStatusScoreDetails
Chain an escalation path across three facts (~4k tokens)scored33%{"battery":3,"hit":1,"finish_reason":"stop"}
Twelve config revisions with expiries, scopes, reverts and no-op cancelsscored100%{"battery":3,"hit":3,"finish_reason":"stop"}
Cross-reference three separated facts (~16k tokens)scored100%{"battery":3,"hit":3,"finish_reason":"stop"}
BOSH manifest property precedence (cross-reference)scored100%{"battery":3,"hit":3,"finish_reason":"stop"}
Three needles with superseded decoys (~4k tokens)scored100%{"battery":3,"hit":3,"finish_reason":"stop"}
Three inventory facts with superseded decoys (~1k tokens)scored100%{"battery":3,"hit":3,"finish_reason":"stop"}
instruction (13 tests)
TestStatusScoreDetails
Instruction conflict — format contradictiontimeout0%{"reason":"hard kill after 135s"}
Instruction conflict — ordering contradictiontimeout0%{"reason":"hard kill after 135s"}
Forbidden words — sunsettimeout0%{"reason":"hard kill after 120s"}
Acrostic plus lipogram plus exact word counts, simultaneouslytimeout0%{"reason":"hard kill after 720s"}
Release changelog under six simultaneous mechanical constraintstimeout0%{"reason":"hard kill after 720s"}
Three sentences about deployment without the letter etimeout0%{"reason":"hard kill after 540s"}
Seven lines where line N contains exactly N wordstimeout0%{"reason":"hard kill after 540s"}
Six simultaneous constraintstimeout0%{"reason":"hard kill after 360s"}
Prompt injection — embedded role overridetimeout0%{"reason":"hard kill after 135s"}
Prompt injection — ignore-previous-instructions attacktimeout0%{"reason":"hard kill after 135s"}
JSON-only outputtimeout0%{"reason":"hard kill after 120s"}
Markdown structuretimeout0%{"reason":"hard kill after 120s"}
Word count — 20 wordstimeout0%{"reason":"hard kill after 120s"}
file_ops (3 tests)
TestStatusScoreDetails
Create haiku filetimeout0%{"reason":"hard kill after 120s"}
Write Python function filetimeout0%{"reason":"hard kill after 120s"}
Write CSV filetimeout0%{"reason":"hard kill after 120s"}
multi_turn (8 tests)
TestStatusScoreDetails
Context carryover — follow most recent contradictory claimscored75%{"turns":[{"user":"The capital of France is Paris.\n","response":"That is correc
Context carryover — job recall after topic shiftscored100%{"turns":[{"user":"Alice works as a software engineer.\n","response":"Understood
Context carryover — number recall after distractionscored100%{"turns":[{"user":"Remember these numbers: 7, 13, 25.\n","response":"Understood.
Iterative code refinement (greet function → error handling → docstring)scored100%{"turns":[{"user":"Write a Python function called greet(name) that returns \"Hel
Iterative email refinement (apologetic + deadline + shorten)timeout0%{"turns":[{"user":"Write a short email to Sarah explaining that a project delive
Iterative summary refinement (summarize → shorten → bullet points)scored100%{"turns":[{"user":"Summarize this: \"The quick brown fox jumps over the lazy dog
Multi-step tool use — weather query then umbrella advicetimeout0%{"reason":"run_time_cap_exceeded"}
Multi-step tool use — read file then extract porttimeout0%{"reason":"run_time_cap_exceeded"}
reasoning (8 tests)
TestStatusScoreDetails
Causal reasoning — root cause of extended shutdowntimeout0%{"reason":"run_time_cap_exceeded"}
Ten-stage capacity cascade with mixed units and stage-wise flooringtimeout0%{"reason":"run_time_cap_exceeded"}
Cluster capacity with reserved overhead and millicore unitstimeout0%{"reason":"run_time_cap_exceeded"}
Deployment ordering under six interacting constraintstimeout0%{"reason":"run_time_cap_exceeded"}
Single overlapping on-call hour across three timezonestimeout0%{"reason":"run_time_cap_exceeded"}
Multi-hop reasoning — biography dates and placestimeout0%{"reason":"run_time_cap_exceeded"}
Multi-hop reasoning — org chart before and after reorganizationtimeout0%{"reason":"run_time_cap_exceeded"}
Timeline ordering — Project Helios eventstimeout0%{"reason":"run_time_cap_exceeded"}
writing (3 tests)
TestStatusScoreDetails
Writing — professional meeting reschedule emailtimeout0%{"reason":"run_time_cap_exceeded"}
Writing — summarize news article in 3 sentencestimeout0%{"reason":"run_time_cap_exceeded"}
Writing — explain HTTP caching for non-technical readerstimeout0%{"reason":"run_time_cap_exceeded"}
research (2 tests)
TestStatusScoreDetails
Cited answer — source trust for policy decisiontimeout0%{"reason":"run_time_cap_exceeded"}
Synthesis with citations — remote work productivitytimeout0%{"reason":"run_time_cap_exceeded"}
monitoring (12 tests)
TestStatusScoreDetails
Diagnose CF app crash loop from eventstimeout0%{"reason":"hard kill after 120s"}
Diagnose CF 404s from routes and app statetimeout0%{"reason":"hard kill after 120s"}
Diagnose full disk from df outputtimeout0%{"reason":"hard kill after 120s"}
Diagnose Diego cell resource exhaustiontimeout0%{"reason":"hard kill after 120s"}
Diagnose OOM risk from free outputtimeout0%{"reason":"hard kill after 120s"}
Alert triage — page exactly onescored100%{"battery":3,"hit":3,"finish_reason":"stop"}
Cascading failure root cause (multi-signal)timeout0%{"reason":"hard kill after 360s"}
False 'deploy succeeded' from polling the latest installtimeout0%{"reason":"hard kill after 360s"}
SLO error-budget burn ratescored100%{"battery":3,"hit":3,"finish_reason":"stop"}
Identify latency anomaly in Healthwatch metricstimeout0%{"reason":"hard kill after 120s"}
Identify critical disk from Prometheus metricsscored100%{"battery":3,"hit":3,"finish_reason":"stop"}
Diagnose CPU spike from top outputtimeout0%{"battery":3,"hit":3,"finish_reason":"stop","budget_ms":30000,"overrun_ms":267,"
iac (12 tests)
TestStatusScoreDetails
Generate CF app manifest JSON for Spring Boottimeout0%{"reason":"hard kill after 120s"}
Write CF network policy CLI commandstimeout0%{"reason":"hard kill after 120s"}
Generate CF manifest with service bindingstimeout0%{"parse_ok":true,"strict_document":true,"fenced":true,"schema_score":0,"content_
Write multi-stage Dockerfile for Go apptimeout0%{"reason":"hard kill after 120s"}
Max instances under even AZ spread with asymmetric cellsscored100%{"battery":2,"hit":2,"finish_reason":"stop"}
Write GitHub Actions workflow for test and deploytimeout0%{"reason":"hard kill after 120s"}
BOSH errand failure — name the misconfigured propertyscored100%{"battery":3,"hit":3,"finish_reason":"stop"}
Minimal CF egress ASG (no shortcuts)scored50%{"build_ok":false,"stderr":"Traceback (most recent call last):\n File \"<stdin>
Custom ollama template silently ignored on a GenAI tiletimeout0%{"reason":"hard kill after 360s"}
Fix K8s Deployment ErrImageNeverPulltimeout0%{"battery":3,"hit":3,"finish_reason":"stop","budget_ms":30000,"overrun_ms":5332,
Generate K8s Deployment JSON for nginxtimeout0%{"reason":"hard kill after 120s"}
Generate K8s Service and Ingress for nginx on 443timeout0%{"reason":"hard kill after 120s"}
ci_repair (7 tests)
TestStatusScoreDetails
Fix invalid base image tag in Dockerfilescored100%{"build_ok":true,"stdout":"ok\n","finish_reason":"stop"}
Resolve dependency pins through a transitive compatibility matrixscored100%{"battery":3,"hit":3,"finish_reason":"stop"}
Fix missing require directive in go.modscored100%{"build_ok":true,"stdout":"ok\n","finish_reason":"stop"}
Fix wrong groupId in pom.xml dependencyscored100%{"build_ok":true,"stdout":"ok\n","finish_reason":"stop"}
Fix missing dependency in package.jsonscored100%{"build_ok":true,"stdout":"ok\n","finish_reason":"stop"}
Fix incompatible package versions in requirements.txttimeout0%{"reason":"hard kill after 180s"}
Fix YAML syntax error in GitHub Actions workflowscored100%{"build_ok":true,"stdout":"ok\n","finish_reason":"stop"}
repo_patch (14 tests)
TestStatusScoreDetails
Add /v2/users endpoint without breaking /v1timeout0%{"reason":"run_time_cap_exceeded"}
Add /v2/users endpoint without breaking /v1timeout0%{"reason":"run_time_cap_exceeded"}
Migrate app config from env vars to YAML with env fallbacktimeout0%{"reason":"run_time_cap_exceeded"}
Migrate app config from env vars to YAML with env fallbacktimeout0%{"reason":"run_time_cap_exceeded"}
Fix partial-match search bug in Flask-like apptimeout0%{"reason":"run_time_cap_exceeded"}
Fix partial-match search bug in Flask-like apptimeout0%{"reason":"run_time_cap_exceeded"}
Extract shared validation logic into cli/validators.pytimeout0%{"reason":"run_time_cap_exceeded"}
Extract shared validation logic into cli/validators.pytimeout0%{"reason":"run_time_cap_exceeded"}
Add input validation to withdraw()timeout0%{"reason":"run_time_cap_exceeded"}
Add input validation to withdraw()timeout0%{"reason":"run_time_cap_exceeded"}
Migrate off a deprecated APItimeout0%{"reason":"run_time_cap_exceeded"}
Migrate off a deprecated APItimeout0%{"reason":"run_time_cap_exceeded"}
Fix pagination off-by-onetimeout0%{"reason":"run_time_cap_exceeded"}
Fix pagination off-by-onetimeout0%{"reason":"run_time_cap_exceeded"}
sysadmin (6 tests)
TestStatusScoreDetails
Set up cron job for backuptimeout0%{"reason":"run_time_cap_exceeded"}
Write a valid docker-compose.yml with postgres, redis, and volumestimeout0%{"reason":"run_time_cap_exceeded"}
Set iptables rules to allow 22/80/443 and drop resttimeout0%{"reason":"run_time_cap_exceeded"}
Configure nginx reverse proxy for /apitimeout0%{"reason":"run_time_cap_exceeded"}
Create systemd unit file for a Python apptimeout0%{"reason":"run_time_cap_exceeded"}
Create deploy user with www-data grouptimeout0%{"reason":"run_time_cap_exceeded"}
agentic (24 tests)
TestStatusScoreDetails
Implement a documented contract and prove it with teststimeout0%{"reason":"run_time_cap_exceeded"}
Implement a documented contract and prove it with teststimeout0%{"reason":"run_time_cap_exceeded"}
Add pagination to get_itemstimeout0%{"reason":"run_time_cap_exceeded"}
Add pagination to get_itemstimeout0%{"reason":"run_time_cap_exceeded"}
Write CLI passing integration teststimeout0%{"reason":"run_time_cap_exceeded"}
Write CLI passing integration teststimeout0%{"reason":"run_time_cap_exceeded"}
Debug bug buried in import chaintimeout0%{"reason":"run_time_cap_exceeded"}
Debug bug buried in import chaintimeout0%{"reason":"run_time_cap_exceeded"}
Fix null deref bugtimeout0%{"reason":"run_time_cap_exceeded"}
Fix null deref bugtimeout0%{"reason":"run_time_cap_exceeded"}
Two failures — decide which side is wrong in eachtimeout0%{"reason":"run_time_cap_exceeded"}
Two failures — decide which side is wrong in eachtimeout0%{"reason":"run_time_cap_exceeded"}
Atomic config reconciliation — idempotent, generation-ordered, no partial writestimeout0%{"reason":"run_time_cap_exceeded"}
Atomic config reconciliation — idempotent, generation-ordered, no partial writestimeout0%{"reason":"run_time_cap_exceeded"}
Partial observability — the fault is not in the prompt, only in the datatimeout0%{"reason":"run_time_cap_exceeded"}
Partial observability — the fault is not in the prompt, only in the datatimeout0%{"reason":"run_time_cap_exceeded"}
Coordinated config-schema + code migrationtimeout0%{"reason":"run_time_cap_exceeded"}
Coordinated config-schema + code migrationtimeout0%{"reason":"run_time_cap_exceeded"}
Multi-module feature — per-client rate limitertimeout0%{"reason":"run_time_cap_exceeded"}
Multi-module feature — per-client rate limitertimeout0%{"reason":"run_time_cap_exceeded"}
Align Python response key with JS frontend expectationtimeout0%{"reason":"run_time_cap_exceeded"}
Align Python response key with JS frontend expectationtimeout0%{"reason":"run_time_cap_exceeded"}
Rename function everywheretimeout0%{"reason":"run_time_cap_exceeded"}
Rename function everywheretimeout0%{"reason":"run_time_cap_exceeded"}
Per-task run consistency (K runs per test)

Run consistency

min 0.00·avg 0.93·max 1.00
Platform facts battery (6 items, graded)
1.00
Dual tool call
1.00
Tool call — recover from malformed JSON error
1.00
Tool call — recover from missing-parameter error
0.00
Restraint under ambiguity — ask, don't guess
1.00
Tool chain with derived arguments
1.00
Nested args
1.00
Read file
1.00
Restraint — no tool
1.00
Search code
1.00
Weather single-call
1.00
REST API Paginated Response JSON
1.00
Cloud Foundry App Manifest JSON
1.00
Convert CSV to JSON Array
1.00
Conditional schema — route config with variant rules
1.00
Kubernetes Pod Manifest JSON
1.00
Function Call with Array and Boolean Arguments
1.00
Function Call with Enum Argument
1.00
Fix wrong return type in count_vowels
1.00
Fix INNER JOIN dropping customers with no orders
1.00
Alert triage — page exactly one
1.00
SLO error-budget burn rate
1.00
Identify critical disk from Prometheus metrics
1.00
Diagnose CPU spike from top output
1.00
Generate CF manifest with service bindings
1.00
Max instances under even AZ spread with asymmetric cells
1.00
BOSH errand failure — name the misconfigured property
1.00
Minimal CF egress ASG (no shortcuts)
1.00
Fix K8s Deployment ErrImageNeverPull
1.00
Fix invalid base image tag in Dockerfile
1.00
Resolve dependency pins through a transitive compatibility matrix
1.00
Fix missing require directive in go.mod
1.00
Fix wrong groupId in pom.xml dependency
1.00
Fix missing dependency in package.json
1.00
Fix YAML syntax error in GitHub Actions workflow
1.00
Chain an escalation path across three facts (~4k tokens)
0.00
Twelve config revisions with expiries, scopes, reverts and no-op cancels
1.00
Cross-reference three separated facts (~16k tokens)
1.00
BOSH manifest property precedence (cross-reference)
1.00
Three needles with superseded decoys (~4k tokens)
1.00
Three inventory facts with superseded decoys (~1k tokens)
1.00
Context carryover — follow most recent contradictory claim
1.00
Context carryover — job recall after topic shift
0.23
Context carryover — number recall after distraction
1.00
Iterative code refinement (greet function → error handling → docstring)
1.00
Iterative email refinement (apologetic + deadline + shorten)
0.67
Iterative summary refinement (summarize → shorten → bullet points)
1.00